Smartphones' accelerometer can track strokes on nearby keyboards | Hacker News Reader

Smartphones' accelerometer can track strokes on nearby keyboards

digitallounge.gatech.edu

55 pointsduck12 comments

Open articleView on HN

I read somewhere that this can also be done with some accuracy by listening to the sounds of keypresses.

Other examples of (mostly) passive information gathering:

- screens can be read via electromagnetic emissions (at least CRTs could, I don't know about LCD displays these days)

- listening in to a room by bouncing a laser off of a window

- wifi snooping, cell snooping (both obvious)

- car tracking via the rfids embedded in some tires

- person and passport tracking via the rfids in the newest generation of passports (and credit cards)

- printer signature tracking of documents - http://news.bbc.co.uk/2/hi/technology/3753886.stm (old, don't know how much it's advanced)

- the wide world of temporal correlation attacks (both on social networks and in the physical world with http://en.wikipedia.org/wiki/Multilateration)

I got a bit sidetracked there. But yea, this is yet another one for the list. Of course, if you've hacked into the phone, you could also try breaking into any wifi networks or, you know, recording conversations.

Also the rather neat Information Leakage from Optical Emanations[1], which demonstrated the ability to read data from networking equipment based on the state of the Rx/Tx LEDs on the front panel. In a lot of cases, the LEDs were being updated for individual bits sent and received.

I can't remember where I came across it, but there was a vendor with an advisory of something like "Mitigation Strategy: Cover Indicator LEDs with light-insulating tape".

Edit: Just remembered about the demo of an attack against signals radiated by (non-wireless) keyboard leads, and their detection and decoding: http://lasecwww.epfl.ch/keyboard/

[1] http://applied-math.org/acm_optical_tempest.pdf (PDF, obviously)

Fascinating preliminary research work. Side-channels in cryptography are infamous but it is clear that the concept of side-channels is a more generic concept about information leaking in ways that aren't typically considered and/or modeled.

The fact that an accelerometer could detect vibrations from typing reminded me of this work on Acoustic Cryptanalysis [1]. It is a fascinating read. The authors found out that the PC Chips M754LMR motherboard had a bank of 1500µF capacitors near the CPU and power connector and of all the possible places, this was leaking information about the CPU's HLT idling state and otherwise. And it turns out that (at a proof-of-concept level) it was sufficient to break crypto implementations.

[1] http://cs.tau.ac.il/~tromer/acoustic/

This seems like pretty much the same thing as the passwords-over-ssh thing from a few years ago. The timing between keypresses is almost as good as the keypresses themselves. And you can get those via an accelerometer on a nearyby smartphone, or you can watch the packetstrem as they tap away on their ssh connection.

Back channels are fun.

Passwords usually aren't sent letter-by-letter as far as I'm aware. Although that's mostly because doing so would make the server side much more complicated than it needs to be.

These sort of attacks are very interesting though. Information introduces all sorts of regularities into the environment and as long as they aren't big enough to disrupt functionality most people never even think about them... For that matter, people seem to have a general problem with thinking about "outside-the-box applications" (probably not the best way to phrase that).

The attack is for when you get a password prompt on the other end, like for su, sudo, or logging into another machine. The password for the original connection is sent as one packet, but the other passwords are normal keystrokes.

Well, in TCP anyway. LAT was a line-oriented protocol. This is why back in the day VMS boxes could support so many more users than Unix boxes - the host didn't have to process every keypress and the network didn't have to send a packet for it, if you were dealing in complete lines. The flipside obv is that VMS needed (very) slightly smarter dumb terminals.

Ah, my mistake. Interesting, thanks!

Ignoring the security implications, this could be a great way to use an old spare keyboard for text entry on your phone. And with some training, it would probably be far more than 80% accurate.

That would be pretty interesting since you don't even need to hit the right key - just the left side of the keyboard or the right.

I wonder I could learn to type by just tilting the phone left right in the correct pattern.

Not to rain on your parade here, but eighty percent accuracy would require every sixth keypress to be backspace. Likely promoting it to the most pressed key.

Obviously, auto-correct mitigates this somewhat but for the amount of frustration and lost time, I'm betting it's probably worthwhile just purchasing a new keyboard.

But what if the keyboard – unpowered and wireless – were designed for this purpose, emitting unique sounds per key, perhaps even outside normal human hearing range?

Reply on news.ycombinator.com